Replace _PyUnicode_READY_REPLACE() and _PyUnicode_ReadyReplace() with unicode_ready()

 * unicode_ready() has a simpler API
 * try to reuse unicode_empty and latin1_char singleton everywhere
 * Fix a reference leak in _PyUnicode_TranslateCharmap()
 * PyUnicode_InternInPlace() doesn't try to get a singleton anymore, to avoid
   having to handle a failure